Development of real-time pedestrian detection method using HOG features /
Pedestrian detection plays a key role in several important applications, such as intelligent surveillance and security, crowd control, and especially in traffic safety for both pedestrians and vehicles. There are many types of features that can be used to represent pedestrians as objects in images....
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Language: | English |
| Published: |
Gombak, Selangor :
Kulliyyah of Engineering, International Islamic University Malaysia,
2016
|
| Subjects: | |
| Online Access: | Click here to view 1st 24 pages of the thesis. Members can view fulltext at the specified PCs in the library. |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Pedestrian detection plays a key role in several important applications, such as intelligent surveillance and security, crowd control, and especially in traffic safety for both pedestrians and vehicles. There are many types of features that can be used to represent pedestrians as objects in images. Using the Histograms of Oriented Gradients (HOG) features is very common in this domain because of its various advantages. However, it suffers mainly from low detection speed that prevents using these features in real-time applications. This is due to the huge number of complex and redundant calculations involved in extraction the HOG features. An attempt to address this problem is proposed in this research, by developing a method to improve the speed of detection while maintaining the accuracy performance as much as possible. The speed of detection has been substantially increased through two procedures: fast extraction of HOG features, and reducing the image rescaling operations that used to build the image pyramid. The first procedure is carried out by speeding up the trilinear interpolation and removing the redundant operations, which involve using the sub-cells method and the concept of reusing the calculated features. The second procedure is done using an approximation technique in order to estimate features in different image scales. The software implementation is written in C/C++ and using OpenCV library (2.4.10), and tested on a PC machine with a general purpose processor (Intel Core i5 M520 @ 2.4GHz) and a sufficient amount of memory (4GB RAM), which is running under Ubuntu Linux 14.10 operating system. The results have shown a significant speed up by nine times over the original HOG algorithm for 640 x 480 images, with a reduction of 2.8% in the overall accuracy performance, using the first procedure only. However, using both procedures increased the total detection speed about 38%, with an approximately reduction of 8% in the overall accuracy performance. |
|---|---|
| Physical Description: | xiv, 109 leaves : ill. ; 30cm. |
| Bibliography: | Includes bibliographical references (leaves 87-90). |